Commit Graph

Select branches

Hide Pull Requests

dsv41-pd

dsv41-pd-visioncp

main

Mono Color

• bec6248272 [Unified Cache]: add glm5.2 per commit ci (#36281) Zhangheng 2026-08-26 10:04:19 +08:00
• 4b2c182d3f [diffusion] feat: support out-of-tree torch.compile backends (#36249) Joe 2026-08-25 21:53:39 -04:00
• 586a50cd96 [AMD][CI] Restore MiniMax-M2.5 4-GPU MI35x nightly job (#36393) 3 people 2026-08-25 18:34:51 -07:00
• 054f485d38 [diffusion] docs: add MiniMax H3 checkpoint format table (#36028) Mick 2026-08-26 09:18:07 +08:00
• 223dfce917 fix: bound CUDA memory for fast image preprocessing (#36295) Mick 2026-08-26 09:02:08 +08:00
• 41e7612dee [Model] Support Nemotron 3.5 Lightning speculative decoding (#36186) Baizhou ZhangandRyan Stewart 2026-08-25 16:43:58 -07:00
• 2d88c79b3e [CI] Add Kimi-K3 MMMU-Pro accuracy coverage (#36284) Baizhou Zhang 2026-08-25 16:33:46 -07:00
• aa718f7343 Refactor HiCache host pool management (#36232) cctry 2026-08-25 16:31:57 -07:00
• 0c42a44cd7 Add Spark3 Model (#35963) 3 people 2026-08-26 07:24:00 +08:00
• f7a56494b1 Fix SWA ownership across grouped frees (#36381) Leon Gao 2026-08-25 14:57:30 -07:00
• 4b4bf3d2a5 [Deepseek-V4] Enable shared-experts fusion on the flashinfer_mxfp4 (trtllm-gen) MoE path (#35505) Junpan Wu 2026-08-26 05:55:25 +08:00
• 6569125e3a [AMD][CI] Add MiniMax-M3-MXFP8 MI35x nightly perf benchmark (#36142) Michael 2026-08-25 14:05:01 -07:00
• edff717ef0 fix(disagg): snapshot affected rooms before iterating outside the lock (#36351) Shangming Cai 2026-08-26 03:31:10 +08:00
• 7ddf92d5f4 fix(disagg): refresh stale prefill bootstrap metadata (#36029) Kevin FlansburgandShangming Cai 2026-08-25 12:29:20 -07:00
• 99c02d71b1 docs(cookbook): use auto parser resolution for Granite 4.2 (#36342) Xinyuan Tong 2026-08-26 01:38:46 +08:00
• 96a73c4e15 [AMD][CI] Adjust MI300 score API performance thresholds (#36290) Bingxu Chen 2026-08-26 00:41:36 +08:00
• e9c9df6a52 [Performance] Tune FlashInfer EXTEND for DP prefill (#36219) YAMY 2026-08-25 08:29:57 -07:00
• b760f7fb19 docs(cookbook): add IBM Granite 4.2 cookbook (#36286) Xinyuan Tong 2026-08-25 22:54:55 +08:00
• 46d9427b91 Fix MXFP8 MoE weight sizing for non-gated models (#36097) elvischenv 2026-08-25 22:09:09 +08:00
• a1f9508dd4 [Bugfix] Fix int32 destination offset overflow in CUTLASS MoE pre-reorder (#35188) Benjamin Truong 2026-08-25 19:56:42 +07:00
• 829138a31e [HiCache] Fix PP inconsistency with HiCache L3 (#22607) (#27010) 5 people 2026-08-25 20:49:57 +08:00
• c3947eeada [diffusion] docs: desktop-safe 24 GB recipe and the DGX Spark tier (#36169) MickandClaude Fable 5 2026-08-25 20:25:17 +08:00
• 0c7ff19e3b [NPU] [CI] Split the base-c-test-acc-2-npu-a3 task into two parts running in parallel (#36257) hhhh1252023 2026-08-25 19:23:20 +08:00
• 443527af0c config: the model-config cache keys on the path the record carried (#36300) Cheng Wan 2026-08-25 03:18:28 -07:00
• 6a81038317 [AMD] Drop redundant FP8 bpreshuffle scale transpose via fused AR kernel (#33021) jacky.cheng 2026-08-25 17:49:57 +08:00
• a618d4c064 [AMD] Add Kimi-K2.7-Code-MXFP4 to cookbook (#36246) 3 people 2026-08-25 17:37:32 +08:00
• 3bc1c580c6 [AMD] Enable draft_extend CUDA graph for HIP DSA backend (#35672) jiaryang 2026-08-25 17:18:08 +08:00
• 8f096b853a [diffusion] fix: crop GLM-Image output to requested size (#33859) 3 people 2026-08-25 16:50:47 +08:00
• 2e3934f4cb [AMD] Enable Mori-EP on kimi-k3 (#35630) Xinyi Song 2026-08-25 01:48:09 -07:00
• d067622820 [AMD][bugfix] Add moe_ep_size/moe_tp_size to the allreduce-fusion gate test stub (#35340) Bingxu Chen 2026-08-25 16:24:52 +08:00
• 191244b3f6 [diffusion] feat: support loading Comfy NVFP4-AWQ text encoders (#36046) Mick 2026-08-25 15:43:30 +08:00
• bf1e03f712 [MegaMoE] Respect padded MXFP8 scale row strides in pre-dispatch (#36237) Lianmin Zhengandyangliu991 2026-08-25 00:27:33 -07:00
• e2b50930b9 [NPU] DeepSeek-V4 adapt sgl-kernel-npu ops (compressor/sparse-attn/sparse-attn-metadata) (#35676) 4 people 2026-08-25 15:13:12 +08:00
• 61b67316d8 [NPU] [Diffusion] Support MiniMax H3 on Ascend NPU's (#33569) Артем Савкинandronnie_zheng 2026-08-25 09:44:14 +03:00
• b7f9fca26e [CPU] Raise mem-fraction-static 0.1->0.2 in intel_amx backend a/b (#36260) MingxuZhandClaude 2026-08-25 14:36:20 +08:00
• f2ef826f0c [NPU]Fix run_lora_a_embedding out-of-vocab token produces wrong embedding. (#32031) qinsir5522 2026-08-25 14:09:25 +08:00
• 7769ff8f1e [DeepGEMM] Deduplicate JIT precompile across local ranks (#36231) Eric ZhangandNikita Shulga 2026-08-25 01:58:03 -04:00
• 2d6c12e2fd [AMD] Don't request the unused softmax LSE in the AITER diffusion backend (#32926) 3 people 2026-08-25 10:57:42 +05:30
• 13d2ee8180 [AMD] Skip shared-KV verify test on ROCm 7.0 CI (#36139) chuyehandchuyeh 2026-08-25 13:20:24 +08:00
• 01db8a7829 [BugFix][Qwen3.8] Support Qwen3.8-MXFP4 DCP by registing Qwen3_5 text-only archs in mamba radix cache whitelists (#35297) Wang, FangYuan 2026-08-25 13:13:59 +08:00
• 9eee990ce1 [AMD] cookbook: add HiCache host-DRAM KV tier for Qwen3.5 MXFP4 on MI355X (#36245) jacky.cheng 2026-08-25 12:55:52 +08:00
• 284ed9d1d3 [diffusion] feat: support LongCat-Image-Edit and LongCat-Image-Edit-Turbo (#35829) 3 people 2026-08-25 12:42:08 +08:00
• 67853c5804 [diffusion] feat: dispatch fp8 companions in mixed NVFP4 checkpoints (#36066) Mick 2026-08-25 11:26:37 +08:00
• 833be86c15 [CP V1 Deprecation 1/5] Migrate tests to strategy-based prefill CP (#36222) Baizhou Zhang 2026-08-24 20:15:34 -07:00
• a0f52eca98 [NPU] Add test for --dllm-fdfo (#33634) longxin9715 2026-08-25 10:58:45 +08:00
• 1fa32d50e1 [XPU] Use SYCL kernels for DeepSeek V4 MHC on XPU (#32166) Cui Lily 2026-08-25 10:27:30 +08:00
• 998eeda0a5 [CI] Register test_dflash_logits at its real cost (#36242) Alex NailsandClaude Opus 5 2026-08-24 18:22:49 -07:00
• 5ffdb02d0c [CI] Cut repeated tokenizer loads, serial subprocesses and a double scan (#36241) Alex NailsandClaude Opus 5 2026-08-24 18:22:43 -07:00
• 68575b23d0 [CI] Stop the config ratchets re-parsing the package on every scan (#36240) Alex NailsandClaude Opus 5 2026-08-24 18:22:32 -07:00
• f8f9226cd2 [diffusion] feat: support component-scoped quantization overrides (#36035) Mick 2026-08-25 09:20:47 +08:00
• ddea7b9156 [diffusion] feat: support mixed Comfy NVFP4 and INT8 layers (#36061) Mick 2026-08-25 09:19:47 +08:00
• d2d8ecea77 [npu] Combine NPU test fixes from #35472 and #34516 (#36180) pllimax 2026-08-25 09:00:40 +08:00
• 91e7e84ee5 [SM120] flash_mla: allocate the page-split buffer outside inference mode (#35116) eeecho 2026-08-25 08:42:23 +08:00
• 0f7ba3d115 [HiSparse] Support hisparse multi-step swap io kernel (#32162) huangtingwei 2026-08-25 08:29:44 +08:00
• 1ec20fd25d [CI] Stop the resolution ratchets re-parsing the whole package (#36235) Alex NailsandClaude Opus 5 2026-08-24 16:55:35 -07:00
• effe0d14d2 [Fix] Keep the MiniCPM-SALA config reads visible to the resolution ratchets (#36178) 3 people 2026-08-25 07:05:09 +08:00
• 3c481b9421 [Benchmark] Add optional steady-state window for serving metrics (#30918) BingjiaWang 2026-08-25 05:49:20 +08:00
• 24bce93c93 [AMD][MORI] Deduplicate CP-replicated state transfers (#36025) AMD-yanfeiwang 2026-08-25 05:43:52 +08:00
• 0d5b5ae620 [AMD] Fix Qwen3.5 MTP dropping fused shared-expert weights (#35719) jacky.cheng 2026-08-25 04:18:42 +08:00
• 0665740953 [AMD][Fix] Route MoRI through the Qwen MoE all-to-all path (#32039) jacky.cheng 2026-08-25 04:04:08 +08:00
• 6e2f87d589 docs: mark Ling-3.0-flash DSPARK verified for all four quantizations on H200 (#36204) Xinyuan Tong 2026-08-25 02:05:32 +08:00
• 5030637c65 [docs] Split the Qwen3.8-27B NVFP4 cells by lm_head precision (#36020) Jimmy Shong 2026-08-24 10:50:33 -07:00
• d10a656ad8 Cleanup duplicate mamba backup helper (#36203) Ke Bao 2026-08-25 00:52:37 +08:00
• 586211bc46 Add PD test for inkling with mxfp8 KV (#35840) Ke Bao 2026-08-25 00:12:05 +08:00
• 54ec2c4699 Fix recurrent state loss on decode retraction (#35957) Ke Bao 2026-08-25 00:11:05 +08:00
• 30f9ed09d1 [diffusion] feat: support loading minimax h3 gguf text encoders (#36055) Mick 2026-08-24 23:00:49 +08:00
• 9b0007ed19 [diffusion] feat: support loading comfy nvfp4 minimax h3 checkpoints (#36044) Mick 2026-08-24 22:57:02 +08:00
• 76d1401881 [diffusion] feat: support mixed w4a4 and int8 checkpoints (#36040) Mick 2026-08-24 20:35:11 +08:00
• e586a6f2c5 Report the whole server's world size in the scheduler's internal state (#35929) fzyzcjy 2026-08-24 20:21:45 +08:00
• 6dd79576cd Expose the declared sglang env vars of a scheduler in its internal state (#35928) fzyzcjy 2026-08-24 20:20:29 +08:00
• c56cee0f80 Support gated launch to defer startup memory allocation (#35927) fzyzcjy 2026-08-24 20:19:45 +08:00
• 3b24d8981b Report per-token weight-version spans in generation meta info (#35926) fzyzcjy 2026-08-24 20:18:52 +08:00
• 981dfa2b83 Make the scheduler track the published weight version (#35925) fzyzcjy 2026-08-24 20:17:50 +08:00
• 02a3dca738 Extract _make_abort_req from the scheduler's abort paths (#35924) fzyzcjy 2026-08-24 20:16:38 +08:00
• a37fdae562 Extract collect_inflight_reqs from abort_request for reusing (#35923) fzyzcjy 2026-08-24 20:15:09 +08:00
• 9856b58de4 [diffusion] feat: support loading serialized fp8 clip image encoders (#36056) Mick 2026-08-24 19:59:35 +08:00
• b4bd5f91ee [diffusion] Fix test_model_fast_paths import after sana_ln_modulate rename (#36175) Hert4andAlex Nails 2026-08-24 18:48:17 +07:00
• bfeae4e79a [diffusion] feat: support loading serialized convrot w4a4 checkpoints (#36039) Mick 2026-08-24 19:13:35 +08:00
• 716a6bf10c feat(humming): support native W4AFP8 checkpoint schemas (#32033) guzekai01 2026-08-24 18:59:27 +08:00
• 21258b7a35 feat(humming): FP8 DeepEP dispatch for humming MoE backend (#31429) guzekai01 2026-08-24 18:59:07 +08:00
• 46b92b22e2 [diffusion] Accelerate LingBot Video RMSNorm in quality=high (#35969) Xiaoyu Zhang 2026-08-24 18:02:00 +08:00
• d8433868ce [AMD][CI] Temporarily bypass local-registry image pulls (#36171) Bingxu Chen 2026-08-24 17:44:53 +08:00
• 092d85eb87 [Feature] Add MiniCPM-SALA support (#30360) 3 people 2026-08-24 17:25:16 +08:00
• d251fa2453 perf(unified-memory): batch lazy-compaction mapping lookup (#34066) SuperSong 2026-08-24 18:14:45 +09:00
• 51b27f747a [diffusion] feat: support loading pruned minimax h3 components natively (#36070) Mick 2026-08-24 16:39:37 +08:00
• adc09a1f63 [diffusion] feat: support loading self-describing quanto int8 encoders (#36052) Mick 2026-08-24 16:37:09 +08:00
• 5081ad5d4e [diffusion] feat: add per-component quantization overrides (#36084) Mick 2026-08-24 16:35:48 +08:00
• 317da0964e [Intel XPU] support prefill only models for xpu (#35072) gaopengffandMa Mingfei 2026-08-24 16:25:47 +08:00
• 7de80e566c [AMD][CI] Name the ROCm Image That Actually Ran in AMD Job Names (#35686) Bingxu ChenandCursor 2026-08-24 16:07:25 +08:00
• 5c4622341f [GDN] remove XPU path of causal_conv1d_fn and causal_conv1d_update (#35775) jiayisunx 2026-08-24 15:59:23 +08:00
• 2070927e00 [Triton] Bound the sliding-window extend-attention KV loop: -86.6% on SWA layers, -9.4% prefill GPU, bit-identical (#34462) Alex NailsandClaude Opus 5 2026-08-24 00:58:52 -07:00
• 7bbd0ddeb5 [AMD] Quark shared-experts gate: recognise a trailing MTP layer (#36124) 3 people 2026-08-24 15:53:52 +08:00
• f464e77d17 fix(mini-lb): forward the flush_cache timeout param to workers (#36150) Shangming Cai 2026-08-24 15:32:57 +08:00
• 666b08b4a5 [ROCm] Extend the gfx950 extend-attention tile to head_dim <= 128: -43% kernel, -14% TTFT, bit-identical (#34461) Alex NailsandClaude Opus 5 2026-08-24 00:29:39 -07:00
• c439e77872 [bugfix] [NPU] fix transpose batch matmul K*B exceed 65536. (#34715) McZyWu 2026-08-24 15:22:59 +08:00
• e28b9cf7b0 Npu single node test timeout config (#36092) pllimax 2026-08-24 14:38:02 +08:00
• c8e1ddc707 [diffusion] feat: cache LoRA-merged weights in files the page cache can hold (#36062) MickandClaude Fable 5 2026-08-24 14:18:55 +08:00
• 9866fe910b [diffusion] Speed up LingBot high-quality VAE decode (#36024) Xiaoyu Zhang 2026-08-24 14:13:03 +08:00
• cc74aba330 [diffusion] Honor XDG cache for model overlays (#36019) Xiaoyu Zhang 2026-08-24 14:11:26 +08:00
• 97b176e64c Support streaming session on NPU (#32597) ming_wang 2026-08-24 14:07:44 +08:00